How to use AI to find *ALL* the literature for your research | A blended approach

Finding “all” the literature in a research field is less about one perfect search and more about stacking methods—using AI for fast, question-driven discovery, then widening the net with author- and citation-based tools, and finally using keyword generation and the “snowball” technique when momentum stalls. The core payoff: researchers can turn an overwhelming, tool-heavy literature hunt into a structured workflow that rapidly gathers papers, surfaces gaps, and keeps searches current.

A first step is semantic searching—starting with a question rather than a rigid keyword string. Tools that accept natural-language queries can retrieve relevant papers and even summarize multiple abstracts at once, giving a quick “top papers” answer and a launchpad for deeper exploration. The practical value is speed and relevance: when a new question appears mid-search, the researcher can immediately query for papers that address it, then follow the summaries into the broader literature.

Next comes seed-paper expansion, where one promising article becomes the anchor for surrounding work. Three tools are highlighted for this purpose: Litmaps, Connected Papers, and Research Rabbit. Litmaps turns a seed article into an interactive citation map that can be sorted by signals like “momentum” (papers being cited heavily), helping identify what deserves immediate attention. Connected Papers emphasizes prior works and derivative works, with a bias toward more recent derivative research—useful for filling in up-to-date coverage quickly. Research Rabbit also builds a network map, but adds multiple navigation paths (citations, references, and author-linked routes), making it strong for “snowballing” through related scholarship—though it can feel complex at first.

Beyond paper-to-paper navigation, the workflow also recommends tracking people. Many fields cluster around recognizable, productive authors and their teams. Checking an academic’s research portfolio and outputs pages can surface the most recent papers and conference presentations, often updated for promotion cycles. Following PhD students, master’s students, and other group researchers is framed as a way to access the lab’s most active work—where new data and publications are generated.

For keyword-driven discovery, the transcript points to using ChatGPT to generate targeted search terms. Instead of guessing what to type into Google Scholar, a researcher can ask for keywords tailored to a specific topic (example given: organic photovoltaic devices) and then run those terms through Google Scholar to uncover papers that might otherwise be missed.

Finally, when the search feels repetitive or stuck, the “snowball method” is presented as a manual reset. Starting from one paper, the researcher repeatedly follows citations and references—opening many tabs, downloading PDFs, and saving references—without trying to filter too early. The goal is coverage through momentum: by exploring who cites what (and what each paper cites), the search can uncover unexpected subtopics and new directions. The overall message is that AI accelerates discovery, but combining it with seed expansion, author tracking, keyword generation, and citation snowballing is what builds a comprehensive literature base.